Network distance prediction for enabling service-oriented applications over large-scale networks

Abstract

Abstract-The knowledge of end-to-end network distances is essential to many service-oriented applications such as distributed content delivery and overlay network multicast, in which the clients are flexible to select their servers among a set of available ones based on network distance. However, due to high expenditure of global measurements in large-scale networks, it is infeasible to actively probe end-to-end network distances for all pairs. In order to address this issue, network distance prediction has been proposed by measuring a few pairs and then predicting the other ones without direct measurements, or splicing the path segments between each pair with the observation. It has been considered important to improve network performance, and enables service-oriented applications over large-scale networks. In this article, we first illustrate the basic ideas behind network distance prediction, and then categorize the current research work based on different criteria. We illustrate how different protocols work and discuss their merits and drawbacks. Finally, we summarize our findings and point out potential issues and future directions for further research.

TSMF: Network Latency Estimation using Matrix Factorization and Time Series Forecasting

“…; it characterises the overall fitness of the model (less is better) and has also been reported in prior work [8]. Consistency is an important aspect introduced in [5]; it outputs a statistic lying between [0, 1], (closer to 1 is better) reporting on whether the implicationd i,j >d u,v ⇐⇒ d i,j > d u,v holds. The implication, denotes that a predicted measurementd i,j should be greater (or less) than another predicted measurement,d u,v if and only if the true measurement, d i,j is greater (or less) than the other true measurement, d u,v .…”

TNDP: Tensor-Based Network Distance Prediction With Confidence Intervals

“…There have been numerous network distance prediction approaches proposed for the various distributed applications with network coordinates [28], [11], [29], [30], [31]. A detailed investigation on them is given in [7] and [4]. On the basis of the embedded network coordinates, these approaches can be divided into two categories: (a) Euclidean embedding and (b) non-Euclidean embedding.…”

“…For instance, a client would enjoy downloading the desired video from many candidate CDN servers that have the highest bandwidth to them subject to the known network distances. However, due to the high cost of network measurements and severe traffic congestion, it is infeasible to always actively probe end-to-end network distances among all the network entities [4], [5], [6]. Accordingly, it is urgent to design alternative approaches to understand the network distances for ever-incresing distributed applications.…”

“…Network distance prediction has been considered as a promising solution to fulfil this requirement [4], [7]. The main idea is to exploit the limited network distances among a small set of node-pairs, in the form of either one-way delay or more often Round-Trip Time (RTT), to predict unknown distances among other node-pairs, where direct measurements are not performed.…”

Network Measurements with Function-as-a-Service for Distributed Low-latency Edge Applications

“…Typically, prediction or interpolation mechanisms help avoid measuring all point-to-point connections between all possible clients and servers. Huang et al [10] provides a general overview and identifies three types of prediction solutions: (i) coordinate-based approaches, (ii) path fitting approaches, and (iii) data-driven approaches. Our approach does not fit this classification, but it is a lightweight method that can assist as an alternative for point-to-point measurements in more complete solutions.…”